Working Backwards:
Amazon's Secret Management System

Bill Carr

Co-author, Working Backwards; Former VP, Digital Media at Amazon

APR 12 2023

The Window

2003–2007: When Amazon Changed Everything

"This one narrow window where all of the products and all of the processes except for one were developed. The period where we went from zero to one to one to infinity."

From 2003–2007: all major innovation happened in 4 years
This coincided with transition from startup to scale: zero to one → one to infinity
When CEO can't be in every meeting: time to invent systems
Jeff's scientific method: experiment, hypothesize, iterate

Amazon's Operating System

What Amazon Actually Invented

Working Backwards (PRFAQ), write the press release first, then build
Single-threaded leader, one owner, one team, full accountability
Input vs output metrics, control what you can, measure outcomes separately
Disagree and commit, discuss hard, then align and move
Leadership principles, real, measurable, reinforced through hiring and promotion

leadership principles

years at Amazon

The real innovationMost companies focus on product. Amazon was equally obsessed with process. That obsession is what made scale possible.

Single-Threaded Leadership

How Amazon Killed Resource Contention

The problem: shared resource pools create constant conflict and coordination overhead
The solution: dedicated resources (reports directly or via dotted line) to one leader
Shift from project orientation (swarm then leave) to program orientation (permanent ownership)
Team owns metrics they can control (page load time, click-through on results)
Removes daily resource fighting; management becomes "how many resources per team" (quarterly) not "which task comes first" (daily)

The trade-off

You lose functional excellence when engineers aren't reporting to a VP of Engineering. Countermeasures: functional communities, guilds, and peer reviews solve this.

The outcome

Clear ownership = clear success or failure. No more "pushing on a string" across blurry accountability lines.

The Method

PRFAQ: The Most Powerful Tool

Press Release FAQ: Write as if product already shipped
Three core sections: problem, solution, benefit statement
Data-rich, factual: No hyperbole; ground claims in numbers
Force rigor: You can't BS your way through a press release

"This is a process to prevent the product development process from becoming where you just get locked in on shipping stuff. Break it into two: what should we build, then how do we ship it efficiently."

The funnel, not tunnelWrite 100 PRFAQs, maybe 20 reach the CEO. Not everything survives, think like a VC. The best ideas win, not politics.

Contrarian Truths

What Amazon Got Right (And Others Didn't)

✗Avoid compound metrics, they hide the truthINSTEAD →✓ Break each metric apart. Manage them individually. Compound metrics obfuscate what actually drives results.

✗Centralized resource pools are efficientINSTEAD →✓ Dedicated resources to single-threaded leaders = speed, ownership, and accountability. Worth the cost.

✗Debate ideas endlessly to get buy-inINSTEAD →✓ Write it down. Many ideas will die on paper (good). Rest go through concentric circles of review, then teams sprint hard.

✗Think about leadership as title and authorityINSTEAD →✓ Leadership principles are the operating system. Hire, promote, and measure against them relentlessly. They're not nice words, they're your OS.

Based on Bill Carr's episode on Lenny's Podcast. All ideas on this page are from the episode.Watch on YouTube Follow @BillCarr89 on X

GO DEEPER IN THE EPISODE

04:26 Amazon’s workplace evolution 09:54 Amazon’s “fitness function”11:44 Single-threaded leadership 18:07 Implementing a program orientation with single-threaded leadership 20:16 The GM model vs. single-threaded leadership 21:31 Functional countermeasures needed for single-threaded leadership 25:22 Embracing the “disagree and commit” principle 30:22 Understanding disagreements 32:41 Deciphering Amazon’s “Leaders are right, a lot” principle 35:25 An explanation of the working backwards framework 41:16 PR FAQ process: Amazon’s innovation engine 44:47 Deconstructing the PR FAQ structure 43:49 The concentric circle model for sharing PR FAQs 44:55 The customer problem-solution statement 47:52 Create a product funnel, not a product tunnel 51:19 How Amazon promotes action vs. talk 54:35 Amazon’s flywheel and input metrics 1:00:51 Signs you’ve got a good input metric 1:04:23 How mistakes can still be made with working backwards 1:06:54 Why disagreements aren’t necessarily signs products will fail 1:08:02 Examples of failed Amazon projects 1:09:55 Cultivating risk-taking and accepting failure 1:13:57 Amazon’s “bar-raiser” practice for hiring 1:18:21 Selecting Amazon’s bar raisers 1:20:41 Advice on implementing practices from Working Backwards 1:23:10 Bill’s work as an advisor 1:26:05 Lightning round

Every sketch here began as a transcript.

Send us yours and we’ll draw it. $1 a transcript.

Sketch mine →

0:00